How to Use Analytics
The revolution in computing power and the creation of inexpensive, highly sophisticated analytics software over the last few decades have made modeling tools widely available. Businesses can utilize it to help them understand their operation on a deep, meaningful level. It can help them understand their business prospects/customers better as well as ensure optimization in marketing, product and/or service prices, and labor management.
The four types of analytics — descriptive, diagnostic, predictive, and prescriptive — exist on a continuum, and they build upon one another. As companies rise up the slope of analytics, the models might get more and more complicated to deliver, but the value of the model to the business increases substantially. Whereas a descriptive model is limited and only explains what occurred in the business in the past, a prescriptive model can be used to predict not only what will happen in the future but also provide reasoning for why it’s happening.
Descriptive analytics can help business users understand what’s going on in website clickstream, while diagnostic analytics can explain customer spending. Predictive analytics can help forecast which marketing offers might attract a customer, while prescriptive analytics can optimize an entire business’s processes. Analytics can streamline business processes, reduce labor costs, increase customer loyalty, and optimize an operation.
To expand on the four different types of analytics:
• Descriptive analytics asks: “What transpired?” A simple scenario would be a marketing agency or your web team tracks a website visitors’ discovery journey on your website. Alongside gaining insights into standard industry KPI metrics such as CTR, session duration, and more.
• Diagnostic analytics poses the inquiry: “Why did it happen?” This type of analytics is slightly more difficult to attain than descriptive, but it does provide more value to the business. An example would be a company utilizing website traffic analytics and using it to try to understand why there was a spike in traffic.
• Predictive analytics gets into the true power and value of analytics, asking the question, “What will happen?” Marketing departments use analytic predictive models to identify which customers may respond to a coupon or online promotion and create unique promo campaigns for those unique customers. This can cut down on marketing costs as well as reduce unnecessary contact with customers who would have no interest in an offer and might even see it as an unwanted intrusion and create a more targeted campaign.
• Prescriptive analytics is the most difficult type of analytics to attain, but it is also the most valuable. It poses the inquiry, ” Can we reproduce those outcomes?”, “What if we try that sales promo again that increased revenue by 10%?” Questions like those are valuable and creating effective sales strategies and identifying revenue opportunities and management for restaurants, lodging, markets, sports, airlines is an examples of prescriptive analytics.
Descriptive analytics can be used for pattern discovery for things like customer segmentation, which can divide customers into their preferred choice of purchases. Market basket analysis is also considered a descriptive analytics procedure. This type of analysis looks at what items customers tend to buy together. Market basket analysis is useful for bundling offers and promotions as well as gaining insight into what might sell well together. Detailed customer purchasing habits can be used to help develop future products.
Diagnostic analytics utilizes techniques such as decision trees, data discovery, data mining, and correlations to try to understand causation and behaviors. Building a decision tree atop a web user’s clickstream behavior could reveal why a customer clicked her way through a retailer’s website.
Predictive analytics uses data mining, statistics, machine learning, deep learning, and modeling to analyze current and historical facts, then make predictions about future events. Predictive analytics models vary, depending on the behavior or event they are predicting. Most produce a score or a rating, with a higher score indicating a higher probability a given event occurs. Events can be anything from predicting the use of a marketing offer to whether a customer is going to churn and leave the company.
Predictive analytics can be used to forecast behavior that can lead to a competitive advantage over rivals. It can include massive amounts of data and variables that are beyond human comprehension. Predictive analytics can build upon descriptive analytics results to give more meaningful business information. Predictive analytics has helped marketers create more successful campaigns, score consumer credit, assess health risks, detect fraud, and even discover new energy sources. Predictive analytics can also be used to improve customer relationship management, collection analysis, cross- and up-sell opportunities, customer retention, direct marketing, fraud detection, product prediction, risk management, and budget forecasting, among many other things.
Predictive analytics utilizes regression, discrete choice models, logistic regression, time series models, survival analysis, classification and regression trees, machine learning, neural networks, naïve Bayes, and K-nearest neighbors, among others.
For marketers, data from past campaigns can be used to generate predictive models that help track actual campaign responses versus expected ones. Additionally, these models can help marketers create upper and lower “control” limits that will automatically alert campaign managers if a campaign is over or underperforming.
Prescriptive analytics aims to predict not only what will happen but also provide reasoning for why it’s happening to attempt to optimize a key metric, such as profit, labor management, and supply chain optimization. Modeling methods include linear programming, decision analysis, influence diagrams, and predictive analytics working in combination with a certain set of rules. Ultimately, a prescriptive model’s output is a decision.
Like predictive analytical models, prescriptive analytical ones can ingest a mixture of structured, unstructured, and semi-structured data, then utilize business rules that can predict a future path, as well as recommend how to exploit this predicted future.
Analytics is becoming integral for business these days, and vendors such as Alteryx, AWS, Oracle, IBM, SAS, SAP, Microsoft, Google, Qlik, and Tableau, are all developing products to help make analytics simple and easy to implement.
Types of Models
• Customer segmentation — Customers are segmented according to various attributes they present to a brand. For example, customer demographic, customer location, customer spend, types of products customer buys, etc., etc.
• Customer acquisition — these models can build upon the results of the segmentation modeling, identifying the likely characteristics of attractive customers.
• RFM — often used in data and direct marketing to analyze customer value. RFM stands for recency (how long has it been since the last purchase), frequency (how often does a customer purchase during a 12-month period), and monetary value (what is the total spend). Each attribute is data mined from the company’s data warehouse. After each category is defined, appropriate categories are defined, and segments are created from the intersection of the values. If there are three categories for each attribute, then the resulting matrix would have 27 possible combinations. Customer types, such as “Big spender but rare visitor”, “rarely visits, spends little”, or “comes every two months, spend moderate amounts of money” can be created to aid the marketing department in their campaigns.
• Propensity to respond — useful for brand marketers, this model provides the theoretical probability that a sampled person will respond to a particular offer.
• Identify when a customer might return — while it might be impossible to know exactly when a customer will return to make a purchase, brand marketers can often make a pretty accurate prediction by looking at their past purchasing behavior.
• Customer worth — understanding the value of a customer by looking at his or her spending over a set amount of time, while taking into account a customer’s life stage.
• Customer churn — Customer retention is key to building and maintaining a business as well as remaining profitable. In today’s big data and AI world, analytics can help businesses predict churn so they can act proactively to prevent it.
• Marketing optimization — Analytical tools like A/B testing can help identify which brand marketing offers work best. A/B testing identifies offers that drive the highest campaign response rates and create the most revenue and profit. Other modeling tools, such as logistic regression, decision trees, random forests, and discriminant analysis, can generate the likelihood of response scores.
Conclusion
Analytics has been around for decades, and modeling tools can be found at most medium and large businesses today. There are four distinct types of analytics, and these can build upon each other, producing more complicated and more valuable models the higher the models go up on the analytics value chain. Descriptive analytics can help business users understand what’s going on in website clickstream, while diagnostic analytics can explain customer spending. Predictive analytics can help forecast which marketing offers might attract a customer, while prescriptive analytics can optimize an entire business’s processes.
